An approach to assess swarm intelligence algorithms based on complex networks
Santana, C; Keedwell, E; Menezes, R
Date: 12 July 2020
Conference paper
Publisher
Association for Computing Machinery (ACM)
Publisher DOI
Abstract
The growing number of novel swarm-based meta-heuristics has
been raising debates regarding their novelty. These algorithms often claim to be inspired by different concepts from nature but the
proponents of these seldom demonstrate whether the novelty goes
beyond the natural inspiration. In this work, we employed the concept of ...
The growing number of novel swarm-based meta-heuristics has
been raising debates regarding their novelty. These algorithms often claim to be inspired by different concepts from nature but the
proponents of these seldom demonstrate whether the novelty goes
beyond the natural inspiration. In this work, we employed the concept of Interaction Networks to capture the interaction patterns that
take place in the algorithms during the optimisation process. The
analyses of these networks reveal aspects of the algorithm such as
the tendency to achieve premature convergence, population diversity, and stability. Furthermore, we propose the usage of Portrait
Divergence, a state-of-the-art metric to assess the structural similarities between networks. Using this approach to analyse the Cat
Swarm Optimisation algorithm, we were able to identify some of
the algorithm’s characteristics, assess the impact of one its parameters, and compare it to two other well-known methods (Particle
Swarm Optimisation and Artificial Bee Colony). Lastly, we discuss
the relationship between the interaction network and the performance of the algorithm and demonstrate the similarities between
Cat Swarms and Particle Swarms.
Computer Science
Faculty of Environment, Science and Economy
Item views 0
Full item downloads 0